KR20120094415A - Precision measurement of waveforms - Google Patents

Abstract

DETAILED DESCRIPTION The present invention provides a machine-implemented method for digital signal processing, which includes conversion of a physical analog signal received from a data store or analog source or captured by a physical analog device. Obtaining a digital waveform signal; Determining from the digital waveform, two or more measuring matrices;
Each measurement matrix consists of a plurality of cells, each cell representing a corresponding time slice and frequency bin, and the signal energy and corresponding amplitude for the frequency bin and time slice. has an amplitude.
In the present invention the maximums occurring simultaneously in time and frequency are identified and an associative maximum matrix called " precision metrics " is constructed to show the simultaneous maximums and the adjacent marked maximums are connected to the subchains.

Description

Precise Measurement of Waveforms {PRECISION MEASUREMENT OF WAVEFORMS}

The present invention relates to time-frequency-amplitude analysis and measurement of waveforms and compound waveforms. In the specification, "signal" is synonymous with the more commonly used word "waveforms" and all have the same meaning. The composite waveform is a multiple waveform mixed with several waveforms. In the present invention, the waveform refers to the audio frequency range but is not limited to any particular frequency range or complexity.

Compound waveforms preferably measure the waveform and the components produced by the various sources accurately.

This is difficult when it includes signals that are generated by fast changes in amplitude and / or other sources that overlap in time and frequency, low energy signal frequencies that are obscured by high energy signals. If these waveforms can be measured and analyzed more accurately, we can greatly improve our understanding of what the waveforms contain and how they are transformed.

 Waveform analysis is traditionally performed in the time and frequency domains. Typically these waveforms are first captured digitally in amplitude in the time domain. At this point, the conversion factor is used to measure the signal and the result is displayed as a matrix. Various techniques have been developed to extract frequency / amplitude information from time series data, but it is still challenging to represent how frequency and amplitude change over time, especially if there are sudden frequency and / or amplitude changes or signals from multiple sources are the same. Challenge is under conditions that occupy and frequency range.

The common transform for obtaining time, frequency and amplitude information is the Discrete Fourier Transform (DFT). Unfortunately there is a trade off between time resolution and frequency resolution from the size of the DFT. The time window observed by the DFT is proportional to the dimension of the DFT. Therefore, a large dimension DFT can observe a larger time window than a small dimension DFT. Larger time windows like this allow large dimension DFTs to respond slowly to dynamic changes.

Conversely, large dimension DFTs divide the frequency range into more granular pieces. The maximum frequency measured by the DFT is half the sampling rate of the digitized signal. The DFT of dimension X divides the maximum frequency range from zero to X / 2 equal to " bin ". Therefore, the size of each frequency bin in a DFT is equal to twice the sampling rate divided by its dimensions.

Thus, higher dimension DFTs have higher frequency resolutions than lower time resolutions. Low dimension DFTs have high temporal resolution, not low frequency resolution. Because of this trade off, practitioners are studying modified DFTs or other alternatives that dynamically and accurately represent time-varying waveforms with good resolution at time and frequency.

Spectrograms, discrete wavelet transforms, Wigner-Ville, Gabor, and windowing techniques are more accurate for dynamic signals that change in time, frequency, and amplitude. Some of the techniques used to attempt to generate time-frequency representations are: Each technique has advantages and disadvantages: computational complexity, artificial / distortion, time complexity and resolution, frequency accuracy and resolution, and amplitude accuracy and resolution.

The present invention is distinguished from prior art transformations that combine multiple FT data called spectral correlations. In the prior art, spectral correlation is that the same magnitude FT is calculated and the FT data is not obtained simultaneously. Therefore, in contrast to the present invention, the spectral correlation may not have the effect and accuracy to be improved simultaneously in time and frequency. The present invention is designed to avoid conventionally distorted information.

The inventors of the present application hold several patents and are incorporated by reference below.

References

Fast Find Fundamental Method, U.S. Patent No. 6,766,288 B1;

Method of Modifying Harmonic Content of a Complex Waveform, U.S. Patent No. 7,003,120 B1;

Method of Signal Shredding, U.S. Patent No. 6,798,886 B1.

Provisional Application No. 61/119198 (filed November 26, 2008)

It is an object of the present invention to provide a machine-implemented method for digital signal processing.

In order to achieve the above object, a machine-implemented method for digital signal processing according to the present invention,

Obtaining a digital waveform signal from a data storage or conversion of an analog signal; Constructing one or more measuring matrices consisting of cells comprising amplitudes at predetermined times and frequencies; Making a maximum in one or more metrics according to the comparison of cell amplitudes; And identifying correlated maxima cells from multiple measurement metrics occurring simultaneously in time and frequency. A new matrix called a Precision Measuring Matrix is created with the correlated maxima indicated.

 (Note: All maxes are local maxima for brevity, and we simply write “maxima.” The max can mean an overall max and is not a problem below.)

Each measurement matrix is generated by iteratively transforming an input signal using a Fast Fourier Transform (FFT). Each transform represents a ratio of time called time slices in the input signal and yields a column of cells, each cell corresponding to a frequency range called a frequency bin. Cells in the measurement matrix with maximum amplitude along each time slice and / or frequency bin are identified and marked. Many methods of identifying the maximum can be used and yield various forms of the maximum. The maxima in various metrics and / or various types of metrics occurring simultaneously in time and frequency are marked in a new matrix called "precision measurement metrics". Adjacent associative maximums in the PMM are linked together in a partial chain. The acquiring, determining, identifying, selecting, marking and linking are executed in one machine.

The maximum cell is identified by comparing the amplitude of the adjacent cell with the amplitude of the maximum cell or by comparing the amplitude of another predetermined cell. If the amplitude of the cell is greater than the amplitude of the predetermined adjacent cell, the cell is determined to be a simple maximum cell. Other forms of maxima include "little brother" maxima, angular maxima, and bugler maxima. If multiple maxima are found, at the same slide and frequency bin (maximum multiple forms within different MMs and / or within the same MM / cell), the maxima occur simultaneously according to a given test criterion, and the corresponding cell in the PMM is represented as an associative max. Can be. In each cell of the PMM, the amplitude is added to the function of the value of the corresponding cell in the MMs.

The following definitions are used herein.

FT: Fourier Transform-How to calculate the amplitude of the spectrum of a waveform

DFT: Discrete Fourier Transform-An algorithm for calculating the spectral amplitude of discrete waveforms. The DFT output can be complex or real amplitude. Many embodiments disclosed herein require real amplitudes. If not described as complex numbers, all references to the DFT relate to a DFT with a real output.

FFT: Fast Fourier Transform-The fast running DFT method is often used synonymously with DFT. In the present specification, DFT and FFT are used interchangeably.

FIR: Finite Impulse Response-FIR is defined as FIR filter. The FIR will generate similar time and frequency information from the FFT with real amplitude output.

IIR: Infinite Impulse Response-IIR is defined as an IIR filter. IIRs will generate similar time and frequency information from FFTs with real amplitude outputs.

Window: The time portion used by Fourier series. The window size in the DFT is known as the dimension of the DFT. For example, if a signal is digitized at 8000 samples per second, a DFT with 4000 dimensions will operate on 4000 data samples.

Window technology: A general DFT method in which not all samples in a window are treated the same. For example, a simple DFT with dimension 4000 simply converts 4000 samples. In Windows technology, 4000 samples could be modulated by weighting the sample in the middle and mitigating the sample in the beginning and in the end. The window technique is designed to reduce the sidelobes / artifacts in the frequency response of the DFT.

dB: decibel-measurement ratio used to measure and calculate acoustic and electronic equipment

dBFS: dB Full Scale-DB proportional to maximum peak level in digital representation

Time slice: A portion of time, a time slice can be represented by an FFT running in the time window of a given data. However, the window will generally be larger than the time slice. The window appears in the center of the time slice. The size of the time slice is determined not by the size of the window but by the space between successive FFT runs. If a new FFT is executed every 8 samples on a digited signal with 8000 samples per second, the time slice is 8 samples (1 ms) wide. The FFT window may be 4000 samples (1/2 second or 500 times slices) wide.

Frequency bin: A small range of frequencies (e.g., 1702-1704)

Cell: A unit of matrix, typically a cell, representing a frequency bin in a time slice and containing an amplitude in dBFS.

MM: Measuring matrix-A matrix of cells that represents the spectral amplitude of an overtime waveform. Measurement metrics are generated by repeated FFTs (or other transforms or equivalents). Each FFT is a column (or row) of cells during its time slice—one cell for each frequency bin. The amplitude in each cell is the amplitude for the frequency bin in the time slice. The cell is appropriately viewed and marked maximum. When processing continuous signals in near real time, the measurement metrics can be infinitely long. In a limited time waveform, the measurement metrics can be finite length.

Maximum cell: A cell that is marked as having one or more types of maximums.

Simple maximum cell: The amplitude of the cell is larger than the adjacent cell. The cell is a time peak simple maximum if the amplitude of the cell is greater than that of the cell in the same frequency bin just before and after within a given time. If the amplitude of the cell is greater than that of the cell in the same time slice at the top and bottom, then the cell is frequency peak simple maximum. A single cell may be a simple maximum in time and frequency. Time peak and frequency peak Simple maximum cells can be marked separately or used synonymously and are simply referred to as "simple maximum cells" or "simple maxima" or "local maxima." Can be written.

Associated maximum "little brother" cell:

The amplitude of the adjacent cell is at the simple maximum's designated threshold and is greater than that of the other side of the cell. If a cell is a time peak simple maximum, the cell proceeds immediately and becomes a candidate of associated maxima at the same frequency that is continuous. In an embodiment if the amplitude in a candidate cell advancing a simple maximum is within 2 dB of the amplitude of the simple maximum and is greater than the amplitude of the cell advancing it, it may be marked as a little brother.

Analogously, if a cell is a frequency peak simple maximum, the cells in the same time slice of the upper and lower singular maximums in frequency are candidates at the associated maxima.

In a preferred embodiment, if the amplitude of a candidate cell of a simple maximum top in frequency is within 3 dB of the simple maximum's amplitude and is greater than the amplitude of its upper cell at frequency, it may be marked as little brother. . The dB thresholds at little brothers adjacent to time do not have to be the same as the little brothers adjacent to frequency. The time and frequency little brother can be marked separately, can be used synonymously, and can be simply marked as little brother. There can be two single cells.

Angle maximum cell: The cell varies more than a given threshold due to the difference in amplitude between the cell and the adjacent adjacent cell.

In an embodiment, if a cell is greater than 4 dB above the cell, it proceeds immediately in time and at a same frequency bin 1 dB less than the successive cells, a 3 dB difference acknowledges it as a time angle maximum. do. The frequency angle maximum is found analogically by comparing the amplitude of the cell with the amplitude of the cell at the same time slice at the top and bottom of the frequency. Angle maxima may be marked separately as frequency angle maxima and / or time angle maxima and may be used synonymously and simply marked as angle maxima. have. There can be two single cells.

Burger maximum cell: The change from the maximum to a single peak is detected by measuring the amplitude, and the maximum cell enters and exits the DFT window. The energy peak in the frequency bin affects the amplitude of all cells in the bin where the peak is within the cell's conversion window. For example, if the transform window is 500 time slices wide, the peak enters 500 times slices before it exists. A burglar maximum can be declared by comparing the increase in amplitude with a 500 timeslice decrease and by comparing the comparison with a predetermined threshold.

The cell is marked in the middle. If the peak is longer than one time slice during the cycle, the energy ramps up above a number of cells, then ramps down to below 500 cells, and in the middle multiple The cell may be marked with a burglar maxima. Therefore, unlike the other types of maxima, a burglar maximum is not detected by comparing a cell with a neighboring cell. Since the window is wider than a single time slice, the amplitude change can be seen in the cell being removed from the cell that can be marked. Also unlike simple, associated and angle maxima, the Buggler max can exist as a time max. There is no analog frequency maximum here.

Partial: a marked maximum cell or cluster of related marked maximum cells (e.g., simple maximum and its little brothers).

Partial chain: A collection of parts that are linked together. The partial chain contains one or more parts of each time slice.

The partial chain can traverse the time slice in a straight line, curve and / or angled line. The partial chain may be determined as a connected cell.

DnMM: Dimension N measuring matrix-

A measuring matrix with N frequency bins. For example, the D500MM is a measurement matrix with 500 frequency bins. Potentially infinite time dimensions are not used as MM.

PMM: Precision measuring matrix-A MM with correlated maxima marked

DnPMM: Dimension N precision measuring matrix

Figure 1 shows a block diagram for explaining an embodiment according to the present invention,
Figure 2 illustrates measurement metrics with a simple maximum marked according to an embodiment of the present invention,
3 shows a possible category of maximum detection according to an embodiment of the invention,
4 shows a comparison result between measurement metrics to find the maximum that generates and generates an associated maximum or precision measurement matrix (PMM),
5A and 5B show a visualization of two measurement matrices obtained by separate FFTs on the same data according to an embodiment of the present invention.
5C and 5D show a visualization of the maxima, which are each obtained by an analysis engine according to an embodiment of the present invention and judged by two measurement metrics,
FIG. 5E shows the precision measuring matrix determined by two measurement matrices, respectively, obtained by an analysis engine determined from the maxima shown in FIGS. 9C and 9D according to an embodiment of the present invention. shows a visualization of the output data).
6 shows an embodiment according to the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings, in order that the present invention may be easily understood by those skilled in the art. . Other objects, features, and operational advantages, including the object, operation, and effect of the present invention will become more apparent from the description of the preferred embodiment.

An embodiment of the present invention is to generate a "Precision Measuring Matrix (PMM)" as part of a new dynamic computer machine that generates accurate amplitude and frequency information in time, said PMM being a new kind of very useful data. It outputs more accurately than that according to the prior art.

Compared with the prior art, the embodiment according to the present invention has high resolution and accurate measurement of the frequency and amplitude immediately in the time series signal. Embodiments according to the present invention are achieved by multiple combinations, measuring metrics of various sizes simultaneously and searching for MMs associated with the maximum. As described above, the smaller-sized transforms provide a better temporal resolution, while the larger-sized transforms provide a better frequency resolution. For this reason, two or more measurement metrics (different in time and frequency resolution) are used to form the overall PMM.

Additionally, measurement metrics (metrics with different transforms or other windowing techniques but with the same dimensions) can be used to provide better resolution and accuracy. Multiple maxima (same in a single MM) can be used to find the associated maximum in the PMM. Adjacent associative maximums may be associated with a chain represented by a partial chain. This connection may be achieved by connecting two largest cells that are adjacent or coincident in frequency and adjacent or equal in time slice.

In the basic embodiment two MMs are created using a single maxima and FFTs with different dimensions and the brother (in time and frequency) is marked with a 2db threshold during the brother. In a specific embodiment all four types of maxima are dealt with and the cells are simply marked “maxima”. The PMM is generated by comparing a cell in two MMs with corresponding time and frequency and marking the maximum in the PMM when the two corresponding cells have a maximum. In PMM the cell is populated with the average of the amplitudes of the corresponding cells in the original MMs.

For example, suppose a digitized signal of 8000 samples per second. In one embodiment, one MM may be generated by performing an FFT of dimension 1000 every 80 samples (every 10 ms). Each FFT represents a long time slice. The resulting amplification of each FFT is recorded as one row in this MM. Cells that become simple maxima and little brothers are marked with maximum cells.

The FFT of dimension N (another similar technique) produces an MM with N / 2 frequency bins. Therefore, the MM generated by the FFT of dimension 1000 is D500MM. The maximum possible frequency is always equal to half of the sample rate-eg a sampling rate of 8000 is 4000 Hz. Dividing the frequency range of 4000Hz into equal parts of 500 yields a frequency bin 8Hz width (0-8Hz, 8-16Hz… 3992-4000Hz). The cells that are the simple maxima and the little brother are marked as maximum cells.

The D2000MM may be generated by performing an FFT of the dimension 4000 every 80 samples. The resulting amplification of each FFT is recorded in a row within the D2000MM. The frequency bin of the D2000MM is 2 Hz wide (0-2 Hz, 2-4 Hz, 1-3 Hz ... 3998-4000 Hz). Cells that become simple maxima and little brothers are marked with maximum cells.

The D2000MM is D500MM, which is exactly four times the frequency. In contrast, the D500MM is exactly four times the D2000MM in time. Specifically, the D2000MM is computed from a time window of 4000 samples (spanning a half-second- 50 rows or time slices) and the D500MM is computed from a time window of 1000 samples (spanning only an eighth-of-a-second-12.5 rows or time slices).

To get the time resolution of D500MM and frequency resolution of D200MM at the same time, we associate two D500MM and D2000PMM is created. Because each frequency bin in the D500MM is four times as wide as in the D4000PMM, it is confusing to identify the bin in one and the bin in the other. To simplify this, the D500MM is expanded and the cell exactly matches that of the D2000MM. To do this we divide each cell into four equal parts on the D500MM. Each cell is divided in each column of the D500MM. A cell of -4-4 Hz is divided into four cells (-2-4 Hz, -4-6 Hz, and 6-8 Hz). Each new cell maintains its amplitude and maximum generation of the original 0-8Hz cell. The other cells are divided in the same way. Therefore, because the columns correspond exactly in time, the cells in the extended D500MM correspond exactly with the cells in the similarly expanded D4000MM and the generation of the D4000PMM is now in progress.

In each cell of the D2000PMM, the amplitude is added as the average of the amplitudes of the corresponding cells in the D500MM & D2000MM. All functions of both amplitudes are allowed and the amplitude of the cell in the PMM is the simple average of the energy of the cell at the two originating MMs. If the corresponding cell is marked as maximum in both MMs, each cell is marked as maximum.

It is generally known how exactly the heat of the MMs corresponds in time. The first column in the D2000MM is generated from samples 1-4000 at the digitized input. The first row of D500MM is generated from samples 1501-2500 to be centered equally (center is exactly 2000.5). Since the time slice is 80 samples wide, the time slice represented by the column can be considered from samples 1961-2040. The second row (time slice) may then be 80 samples. The second column of D2000MM will be generated by the FFT in samples 81-4080 (centered at 2080.5). And the D500MM will be generated in samples 1581-2580 (also centered at 2080.5). The time slice may be determined as samples 2041-2120 (again centered at 2080.5). Time slices do not overlap.

Other changes may be made in other embodiments of the invention. The time maxima may drop and frequency maxima may be used. Conversely, time maxima may be used. The brother threshold can be changed from 2dB to another value or the brother can be totally eliminated. Two MMs can be created and connected. When two or more MMs are used, the agreement between all MMs is not necessary universally.

For example, if five different measurement metrics are used, five to three at the same time and frequency are required to identify the associated maximum. The associative maximum is limited to the maximum at the highest dimension MM and is performed concurrently at the maximum at the other MM. Criteria for associative maximums vary by application or signal characteristics and include agreement at two or more maximum times and frequencies. Also, a correlated maximum may be determined differently for different frequency ranges. For example, at lower frequencies more dimension measurement matrices will be needed for increased frequency resolution. Conversely, smaller dimension measurement metrics are more smoothly at higher frequencies.

Combinations of the various variations can be used in other embodiments. Window technology may also be used. In another embodiment, multiple MMs are generated using the same dimension FFT, unlike other windowing techniques. Maxima is compared to work the association maximum and generates a PMM. In another embodiment multiple MMs are created and use different dimension and different windowing techniques. For example, the D500MM and D2000MM can be created without windowing technology, while another D2000MM is created with a Hamming window. These three MMs are associated to generate a PMM. Possible combinations are endless,

In another embodiment, a single MM is generated. In a specific embodiment, the D4000MM is created without windowing technology. Frequency peak simple maxima are identified and marked. Also the Buggler max is identified and marked. All cells that are marked with frequency peaks and Buggler max are the associative max and PMM is generated with the associative max marked.

Other changes may be made in other embodiments of the invention. The MM may have a dimension different from 4000. Deconvolution and / or windowing techniques can be used to extend the frequency resolution of the MM. Little brother can be allowed. The younger brother can produce a Buzzler maximum characteristic function instead of a constant of dB. In particular, a "strong" Buggler maximum (showing a larger slope as a signal peak enters and out of the window) can trigger a different brother threshold than a weak Buggler maximum.

In another embodiment, the cells of MMs can be added by complex numbers, using a full complex FFT. The maxima identification may be based on the amplitude and phase of the cell. For example, the little brother and angle maxima thresholds may include phase components. Larger or smaller phase shifts can be maximum.

In another embodiment, some or all of the FFTs are replaced by multiple IIRs or FIRs. Each frequency bin uses a separate FIR or IIR. A common technique using filters is known to identify the frequency spectrum instead of actually converting from one domain to another, and this method is concentrated on being more computational.

Time-delay compensation may be used to adjust group-delay variations in the filter's frequency response to centrally time-slice within the MMs. MMS generated from FIR or IIR is used the same as MMS generated from FFT.

Specific embodiments of the present invention may combine all of the foregoing, such as up to multiple forms in multiple MMs with different dimension FFTs and / or other transform and / or other windowing techniques and / or banks of FIRs or IIRs. The maximum is associated to correspond with a frequency bin and a time slice. When using FIRs or IIRs, longer time slices can be used to reduce the computational load. In this case a single column (time slice) in the MM generated by FIRs or IIRs is associated at a time with two or more rows in the MM generated by the FFT. To achieve a simple association, matrices can be extended in time analogously to how the metrics extend in frequency-or the association can be made without device. The identification of the maximum cell may also be described as "marking", which is not indicated in the embodiment. The method of tracking which cell is the maximum is not used. In a specific embodiment the separation parallel matrix (or array) includes maximum designations. Therefore, the MM or PMM may be referred to herein as a set or array of parallel metrics that collectively includes amplitude and maximum information, rather than as a single metric by characters in the machine.

Figure 1 shows a block diagram for explaining an embodiment according to the present invention. The analog signal receiving apparatus 101a captures a physical analog signal. Therefore, physical analog devices can be considered. The analog signal receiving apparatus 101a may be, for example, a tape player, a microphone, a video camera or another transducer for playing a tape. An analog-to-digital (A / D) converter 101b converts a physical analog signal into a digital format. In an embodiment A / D converter 101b is installed in a digital audio machine that provides a digital audio workstation DAW. The digital format is generated using 16 bit quantization and a sampling frequency of 8 KHz. Each bit of the quantization provides about 6 dB of dynamic range. Therefore, the 16 bit quantization embodiment leads to an overall dynamic range of about 125 dB, which is sufficient for audio applications. Other quantization options are available and include, but are not limited to, 32 bits and 24 bits. The quantization can be a single bit using sampling and sigma-delta modulation as is commonly used in devices such as CD-ROMs. The analog signal receiving device 101a and the A / D converter 101b may be the same device or a separate device.

In a preferred embodiment, instead of providing a digital signal through the analog signal receiving device 101a and the A / D converter 101b, the digital signal is provided by a data storage device 102 that stores a digital signal. Can be. The data storage device 102 is a type of storage device such as a hard disk drive, a network access storage device, or a flash drive. However, the data storage device 102 is not limited to certain embodiments. The data storage device 102 may include a conventional data storage device or a developed device without departing from the spirit of the present invention.

Many forms of signals may be processed by specific embodiments. For example, the digital signal can be an analog signal digitized from the audio spectrum, video spectrum or other spectrum or can be in digital form.

In blocks 103 to 107 the digital signal is received at the analog signal receiver 101a and the A / D converter 101b or received from the data storage 102 and is converted to generate measurement metrics. The transformation may be an FFT.

For example, at blocks 103-107, an FFT is performed on digitized physical analog signals having different dimensions and corresponding spectral resolution. The function of the conversion can be achieved by an IIR bandpass filter and a bank of FIRs with a center frequency properly spaced for the spectrum of interest. Typically two or more transformations (or filter bank sets) are performed. Figure 1 shows five transformations although certain numbers are shown.

In blocks 108-112, the maximum cells in each measurement matrix are identified by various methods that compare with the amplitude of the cells in each measurement matrix. When processing measurement metrics for real-time waveforms, one column or several columns can be processed at one time. Some data is retained from the heat processed prior to the driver compared to the heat being processed.

In block 113, the identified maxima in each measurement matrix are compared in time and frequency to find the association maximum. The comparison may include the same type or different types of maxima. In a specific embodiment the comparison includes the same comparison / block (e.g. block 108) and a different form of maximum. For example, the blocker maximum cell at block 108 is compared to simple and associated with the maximum cell via blocks 108-112.

In block 114 precision measurement metrics are formed. In block 113, the identification position of the time slice coinciding with the frequency bin forms an associative maximum and is marked in the PMM. In PMM, the amplitude of a cell can be added as a function of the amplitude contained in the corresponding cell of the original measurement matrix (eg mean, weighted average, median). This function can be simple (using the amplitude from one of the original metrics) or complex (sensitive to whether the cell within the original matrix is the largest and the largest form).

In a preferred embodiment the output device 15 may be a data storage or transmission device or a visualization device having a user interface for visualizing the measured signal. The visualization device may be, for example, a display device or a printing device. For example, the display device includes a liquid crystal display device and a projection monitor. For example, the printing apparatus may include a toner mounting printer, a liquid ink jet printer, an inkless printer, or the like. An intermediate result of the embodiment according to the present invention may be provided to the output device 115. In a specific embodiment, the output device 115 may be a transmission device for transmitting the PMM to another device. These results can be used by other devices or signal processors to modify the original waveform. In an embodiment the output device comprises a data storage device and a visualization device and the signal processor can be manually adjusted to achieve the desired result.

FIG. 2 shows the measurement metrics generated by blocks 103-107 of FIG. 1 with simple maximum cells marked by blocks 108-112 of FIG.

For example, columns are timeslices and rows are frequency bins. The number in each cell is the corresponding amplitude measurement. The frequency peak simple maximum cells are marked with dark borders, and the time peak simple maximum cells are marked with a gray background.

For example, frequency peak simple maxima are cells are cells in a time slice in a measurement matrix whose amplitude exceeds the amplitude of two adjacent cells at frequency. In Fig. 2, since cells have amplitudes greater than adjacent cells 160,656.3 and 160,687.5, cells 160,671.9 are marked with frequency peak simple maximums (dark boundaries). Similarly, a time peak simple maximum cell is a cell with a local maximum amplitude in the frequency bin. Since the cell in Figure 2 has an amplitude greater than adjacent cells 158, 671.9 and 160, 671.9, cells 159, 671.9 are marked with a time peak simple maximum (shaded background). 3A and 3C show signal strengths of time peak simple maximum and frequency peak simple maximum, respectively.

My brother is not marked in Figure 2, but we can see where it is. You can assume a 3dB threshold for your brother. Cells 160 and 671.9 are marked with dark borders as frequency peak simple shift cells. Adjacent cells 160, 656.3 and 160, 687.5 are candidates for associated maximum cells and cells 160 and 656.3 are close enough within the amplitude to impart to the federated maximum cell (within 3 dB). Since the amplitude of the cell is greater than that of cells 160 and 640.6, the cell can be a brother. The same decision may be made with the federated maximum cell in the time slice. 3b and 3d show the signal strength of the frequency peak simple maximum with the brother and the time peak simple maximum with the brother.

3 shows a possible category of maximum detection according to an embodiment of the invention. Fig. 3 (A) shows the frequency peak simple maximum, Fig. 3 (B) shows a brother with an amplitude almost equal to the frequency peak simple maximum, and Fig. 3 (C) shows the frequency peak simple maximum. 3 (D) shows a time peak simple maximum, and FIG. 3 (A) shows a brother having an amplitude almost equal to the time peak simple maximum.

Fig. 3E shows the maximum angle in the time slice. The third point is the frequency angle maximum. Its amplitude is slightly lower than the amplitude of the left point by a significant amount. If the difference with the point on the left is sufficiently greater than the difference with the point on the right, it is the angle maximum. For example, we can suggest that the angular maximum threshold is 3dB. If the third point is 5 dB larger than the point on the left and 1 dB smaller than the point on the left, the difference (4 dB) exceeds the threshold and it is an angular maximum. In MM the cell can be marked with maximum angle. It can be seen that the threshold difference does not need to be a constant dB value, but can be a function of frequency and amplitude. The angular maximum cell can be an amplitude branch cell from an expected convexity. The analog time angle maximum can be defined as a change in time within the frequency bin.

Fig. 3f shows the signal peak entering and exiting the conversion window, showing the maximum detection of the Burger. The conversion window is wider than the peak, and when the peak enters, the amplitude rises like the peak, while the peak is inside the window, while maintaining a constant, when the peak exits, it falls. . The actual width of the peak can be obtained from the length of the lamp. This rise and fall time difference and period limit the Burger maximum.

FIG. 4 illustrates the correlation of two measurement matrices as performed in block 113 of FIG. The wider frequency bin is at the lower dimension measurement matrix (eg a D500MM). The maximum in the MM is shown by the horizontal line. The narrower frequency bin is in a measurement matrix whose height is eight times the dimension (eg a D4000MM). The maximum in the MM is shown by the vertical line. The correlated maxima (maximum in corresponding MMs in time and frequency) are shaded. This may be the maximum in the PMM of block 114 of FIG.

5A and 5B show a visualization of two measurement matrices obtained by separate FFTs on the same data according to an embodiment of the present invention. The measurement metrics of FIG. 5A are generated using an FFT of 512 sample points (eg, D256MM), and the measurement metrics of FIG. 5B are generated using an FFT of 4096 sample points (eg, D2048MM). The two measurement metrics are output from blocks 103 and 104 of FIG.

5C and 5D show visualization of the maximum cell as determined by the two measurement metrics of FIGS. 5A and 5B. The maximum cells of FIGS. 5C and 5D are output by blocks 108 and 109 of FIG. 1, respectively.

FIG. 5E shows a visualization of the Precision Measuring Matrix (PMM) determined from the associated maximums shown in FIGS. 5C and 5D. The PMM is obtained at block 114 of FIG.

6 shows an embodiment according to the present invention. The input signal 1005 is a digitized version of the physical analog signal captured by the analog device and the digits 101a and 101b, and is obtained from the data storage device 102. The signal processing engine 1001 processes the input signal 1005 and includes a measurement engine 1002, a marking engine 1003, and a comparison engine 1004. The measurement engine 102 generates an MM from the input signal 105. The marking engine 1003 analyzes and marks MMs. The marking engine 1003 identifies and marks each MM (executing blocks 108-115 of FIG. 1). The comparison engine 1004 identifies the maximum cell in the matrix with the corresponding time and frequency and generates a PMM with the identified correlated maxima marked.

In some embodiments of the present invention, signal processing module 91001 may include one or more digital processors. One or more processors are in communication with a data storage device, for example, a data storage 102 that stores software and / or data used by one or more of the digital processors of FIG.

The measurement engine 1002, the marking engine 1003 and the comparison engine 1004 may be executed by a computer having at least one central processing unit (CPU) and a plurality of memories. The measurement engine 1002, the marking engine 1003 and the comparison engine 1004 may include a field programmable gated array (FPGA), a digital signal processing chip (DSP), graphics A graphic processing unit (GPU), an application specific integrated circuit (ASIC), or the like. The measurement engine 1002, the marking engine 1003 and the comparison engine 1004 may be executed in one or more engines and may be stored in a computer readable storage device.

Output data 1006 is a precision measurement matrix. The output device 1007 stores, displays and / or transmits output data.

The invention can be embodied by software in a digital computer machine. The invention is implemented in a single processor with a single core, a single processor with multiple cores, or multiple processors with a single or multiple cores. The software executes blocks 103-112 or a combination thereof.

The invention may be embodied using generic or specific digital signal processor chips. The present invention can be embodied in dedicated hardware. The invention may be embodied in one or more circuit boards, for example boards comprising digital signal processor chips mounted in a computer or other device.

In specific embodiments of the invention, the matrix data may be replaced by other data structures that may provide surprisingly the same functionality. Such data structures may be organized by fields. The data includes but is not limited to a sparse matrix, a linked queue. For example, the identification maximum cell may be represented by a sparse matrix, and the identification chain of the cell may be represented by a linked queue.

Specific embodiments in accordance with the present invention may be implemented using parallel computing hardware. In an embodiment blocks 103-107 and / or blocks 108-112 of FIG. 1 may be implemented in parallel hardware. The main memory of a parallel computer can be shared memory (shared between all processing elements in a single address space) or distributed memory (each processing element has its own local address space).

The present invention provides a signal time-variance analysis that includes audio, audio spectra analyzers, industrially available signal analyzers for all frequency ranges, data compression, digital cochlea for hearing impaired, "voice print" information, and the like. ), But not limited to.

The information developed by the present invention can be used to provide new and improved inputs to signal processors, analyzers, algorithmic systems, and organizations.

Records of related partial chains performed on the time-frequency plane along with amplitude or interpolated amplitudes along each chain provide a rich, desired signal source ( Clarify the recording of components of particular signal sources. When the partials are linked to a partial chain in the PMM visual display, the signal previously hidden by the strong signal can be measured and visualized.

For reference, the specific embodiments of the present invention are only presented by selecting the most preferred embodiments to help those skilled in the art from the various possible examples, and the technical spirit of the present invention is not necessarily limited or limited only by the embodiments. In addition, various changes, additions, and changes are possible within the scope of the technical idea of the present invention, as well as other equivalent embodiments.

Claims (25)

In a machine-implemented method for digital signal processing,
Obtaining a digital waveform signal from conversion of a physical analog signal received from a data storage or an analog source or captured by a physical analog device;
Determining from the digital waveform, two or more measuring matrices;
Each matrix consists of a number of cells,
Each cell represents a corresponding time slice and frequency bin, and has an amplitude corresponding to the signal energy for the frequency bin and time slice,
Identifying the maximum cells in each measurement matrix;
Identifying correlated maxima cells from multiple measuring matrices occurring simultaneously in time and frequency;
Constructing a precision measuring matrix depicting the associative maximum cell; And
Storing, transmitting and / or displaying the precision measuring matrix;
The acquiring, determining, identifying, selecting, storing, transmitting and / or displaying may comprise one or more machines. Machine-implementing method for digital signal processing, characterized in that executed using. The method of claim 1,
The maximum cells in each measurement matrix are identified as simple maximum cells according to a local maximum amplitude. The method of claim 1,
If the cell is a time peak simple maximum cell, the cell is determined to be a local maximum. The method of claim 1,
If the cell is a frequency peak simple maximum cell, the cell is determined to be a local maximum. The method of claim 1,
A machine-implemented method for digital signal processing, wherein the associated maximum cells are determined to be the maximum cells. The method of claim 1,
A machine-implemented method for processing digital signals, characterized in that angle maximum cells are determined to be maximum cells. The method of claim 1,
Machine-implemented method for digital signal processing, characterized in that the Burglar maximum cells are determined as the maximum cells. The method of claim 1,
The various types of maxima are judged to be identical and the identified cells are all identified as maximum. The method of claim 1,
The various types of maxima are not judged to be the same, and each type of maxima is separately tracked separately. . The method of claim 1,
The groups of types of maxima are defined, the types of maxima within each group are judged to be the same, and the different groups of types are judged to be the same. Machine-implementing method for digital signal processing, characterized in that the maximum is tracked separately according to the group. The method of claim 1,
For a corresponding time slice and a corresponding frequency bin, if a majority of the cells of the measuring matrices is a maximum cell, a correlated maximum Is declared and the corresponding cell in the correlated maxima matrix is identified as a correlated maximum. The method of claim 1,
For a corresponding time slice and a corresponding frequency bin, if all of the cells of the measurement matrix are maximum cells, a correlated maximum is declared and the corresponding cell in the associated maximum matrix (the corresponding cell in the correlated maxima matrix) is identified as a correlated maximum. The method of claim 1,
For a corresponding time slice and a corresponding frequency bin, if a given combination of two or more cells of the measurement metric is a maximum cell, a correlated maximum is declared and precision The corresponding cell in the precision measuring matrix is identified as a correlated maximum. The method of claim 1,
An associative maximum is declared and based on the function of the cell of the measurement matrix in a corresponding time slice and a corresponding frequency bin, the corresponding cell in the precision measurement matrix is identified as the associative maximum Machine-implementation method for digital signal processing by means of The method of claim 1,
The amplitude of a cell within the precision measurement matrix is populated by the function of the amplitude of the corresponding cell of the measurement matrix and the function of the amplitude when the cell is the maximum cell. . The method of claim 1,
Adjacent maxima within the precision metrics are connected in a partial chain. Machine-implemented method for digital signal processing. In a machine-implemented method for digital signal processing,
Obtaining a digital waveform signal from conversion of a physical analog signal received from a data storage or an analog source or captured by a physical analog device;
Determining from the digital waveform, two or more measuring matrices;
Each matrix consists of a number of cells,
Each cell represents a corresponding time slice and frequency bin, and has an amplitude corresponding to the signal energy for the frequency bin and time slice,
Identifying two or more maximum cell types in each measurement matrix;
Identifying a correlated maxima cell type in multiple measuring matrices that occur simultaneously in time and frequency;
Constructing a precision measuring matrix depicting the associative maximum cell; And
Storing, transmitting and / or displaying the precision measuring matrix;
The acquiring, determining, identifying, selecting, storing, transmitting and / or displaying may comprise one or more machines. Machine-implementing method for digital signal processing, characterized in that executed using. 18. The method of claim 17,
The multiple maxima used to identify the associated maxima include a burglar maxima and at least a simple, angle or associated maxima. -How to run. 18. The method of claim 17,
And said identified correlated maxima-cells have a Burglar maximum and a different type of maximum. 18. The method of claim 17,
If the cell is a frequency peak simple maximum, the cell is determined to be a local maximum. 18. The method of claim 17,
If the cell is a time peak simple maximum, the cell is determined to be a local maximum. 18. The method of claim 17,
If the cell is a frequency peak simple maximum or a time peak simple maximum, the cell is determined to be a local maximum. 18. The method of claim 17,
A machine-implemented method for digital signal processing, wherein the associated maximum cells are determined to be the maximum cells. 18. The method of claim 17,
A machine-implemented method for processing digital signals, characterized in that angle maximum cells are determined to be maximum cells. 18. The method of claim 17,
Adjacent maxima within the precision metrics are connected in a partial chain. Machine-implemented method for digital signal processing.

Images